Lifted sequence coercions

[treeowl]

Currently, you use data Any a where Any :: a b c -> Any a. This has a potential efficiency problem because of the extra constructor. I'm also not sure if it's entirely safe. As best I can tell, the officially sanctioned way to do this sort of thing is with the somewhat magical GHC.Prim.Any :: *, which is designed as a "safe" intermediary for unsafeCoerce. To use it, you'd just make a Seq Any or similar, and unsafeCoerce both on the way in and on the way out. GHC seems to promise not to play games with this type.

newtype FakeSequence a b c = FakeSequence (Seq Any)

[atzeus]

Where do I use Any? Cheers!

[atzeus]

O wait in LiftSequence :)

[atzeus]

Yep this is definitely an improvement! I'll change it. Thanks!

[treeowl]

I guess you'll want a kind annotation there. Probably something like

newtype FakeSequence (a :: k -> k -> *) (b::k) (c::k) = FakeSequence (Seq Any)

[treeowl]

Any progress on this? Should I open a pull request?

[atzeus]

Fixed by #2ab000be1ddc49a2ec36afaaa58485df56aef654

[treeowl]

I just looked at this again, and was surprised to see unsafeCoerce# rather than the "safer" unsafeCoerce. Why is that primitive needed here?